SedStrat Analytics

Machine Learning implementation in

Reservoir Modelling
July 2020
Please direct your inquiries to: epo@[Link]

epo@[Link]
 Number of Wells: 3032
Analysis to do: -
- Well data QC - 2 hr* 3032 = 6,064 hr
- Well correlation - 6 hr* 3032 = 18,192 hr
- Petrophysical analysis - 3 hr* 3032 = 9,096 hr
- Facies identification - 6 hr* 3032 = 18,192 hr
51,544 hr
~6,444 working days
~19.5 working years

Automate them with With more geologist,

MACHINE LEARNING!! lesser working time
Done in 10 minutes using basic With 5 geos:
Parameter..!! ~19.5 / 5 = 3.9 years

Result: 89% accuracy in Facies

With more geologist
Prediction
subjectivity introduced
More time for interpretation!
epo@[Link]
 Who We re...
SedStrat is a group of Geologist that
capable of reading digital data, and
passionate in automated geological
interpretation using open-sourced Epo Prasetya Mohammad Naufal
Yogi Pamadya
analytics toolkit Kusumah Aviandito Hadiyan

Geologist Geologist Geologist Geologist

Researcher @ Data @ Gojek Data @ LinkAja Data @ Tokopedia

Universitas
Pertamina

University of University of University of University of Salford

Manchester Queensland Manchester Petroleum and Gas
Petroleum Geoscience Sustainable Energy Petroleum Geoscience Engineering graduate
graduate graduate graduate
Data Engineering
Ex-VICO Geoscientist Ex-Halliburton Ex-Pertamina experiences in
Petrophysicist Geoscientist Indonesia’s leading
tech companies

epo@[Link]
 Outline

● Intro
● Automation in Exploration
● Automation in Development
● Demo and Showcases

epo@[Link]
 Why Automate?

Without automation: With automation:

- Data QC (30%) - Setup ML Environment (20%)

- Geological Work - Data QC (5%) Machine Learning
- Routine Repetitive work (30%) - Geological Work Reduced work by total
of 50% Geologist
- Interpretation (20%) - Routine Repetitive work (5%) Work
- Presentation (10%) - Interpretation (50%)
- Presentation (20%)

epo@[Link]
 Data Analytics

Data Information Implementation

Cognitive
Process
Human Cognitive AI
Assessing Create
Qualitative Quantitative
Scoring Draw Small task BIG task
Limited Working stamina Unlim Working stamina

Interpretation etc.. Easy to Train VERY HARD to Train

Geoscience domain knowledge is required in every aspects of the project

epo@[Link]
 Machine Learning

Data ML Information Implementation

Data
Data prep Feature Selection Algorithm Tuning Scoring
preprocessing
Optimization

Machine Learning

Deep Learning

Geoscience domain knowledge is required in every aspects of the project

epo@[Link]
epo@[Link]
 Exploration Stages for Hydrocarbon
Play Based Exploration
Play Drillable
Basin Prospect
Identification Target
Regional tectonic and Prospect ranking by:
Sedimentation reconstruction Is a prospect generation method that describe exploration potential and - Risks
exploration risk for each identified play. The general steps required to generate - Resource
prospect using PBE are: - Economic Model
1. Play definition
2. CRS map for each Play
a. Source rock potential map
b. Reservoir potential map
c. Seal potential map
d. Overburden presence
e. Maturity potential map
f. Migration pathway
g. Presence of Fault
3. Stack all of CRS map for each play to identify Sweet Spot

ML, AI, and DS might be able

to quicken some routine
process. Allowing the
geologist to work on
interpretation more.

epo@[Link]
 Discrete Processed
Information

Auto CRS MAPS

Well Log Reservoir
Petrophysics

Reservoir map

Seal
Seal map

Map Making Prospect

Geochem Passay Source Rock Source rock map Overlay
Algorithm “Sweet Spot”

SVM probability
Maturity Profile SIS
Maturity Map
and Depth Krigging?

Migration map

Data
Well marker +
Auto Correlation
age
Algorithm
Simple Burial
History
NCT Prediction Intermediate
products

Depth Structure Drainage Pattern Migration

Map Reconstruction Pathway

epo@[Link]
 Discrete Processed
Information

Auto CRS MAPS

Well Log Reservoir
Petrophysics

Reservoir map

Seal
Seal map

Map Making Prospect

Geochem ML Geochem Source Rock Source rock map Overlay
Algorithm “Sweet Spot”

SVM probability
Maturity Profile SIS
Maturity Map
and Depth Krigging?

Migration map

Data
Well marker +
Auto Correlation
age
Algorithm
Simple Burial
History
NCT Prediction Intermediate
products

Depth Structure Drainage Pattern Migration

Map Reconstruction Pathway

epo@[Link]
 Development Stages for Hydrocarbon
Second
Discovery Delineation Ramp Plateau Abandonment
Options

● Infill drilling ● Infill drilling ● EOR drilling

Operation ● Water Flood Well
● Workover

Resource ● Resource Estimation ● Reserve Estimation ● Upside Potential

Management Reserve Estimation

Portfolio ● Reservoir Mapping ● Upside Potential ● Bypassed Zone

● LCLR
Management ● Unconventional
Prod Rate

Time

epo@[Link]
 Operation Data streaming -> Interpretation on the fly
● Drilling anomaly detection
● Pressure detection
Formation tester
● Pressure measurement / modelling
● Compartmentalization*
● Fluid analysis

Resource Initial Reservoir Modelling:

● Petrophysical analysis
Management
● Well correlation
● Facies identification
● Facies mapping
● Property mapping
● Resource / reserve estimation

Reservoir model Update

Reservoir property Estimation:
● Rock typing
● J-function / Saturation Height Model

Portfolio ● Reservoir Mapping

● Upside Potential
Management
● LRLC
● Bypassed Zone

epo@[Link]
 Facies Labeling

Operation Data streaming -> Interpretation on the fly

● Drilling anomaly detection
● Pressure detection
Formation tester
● Pressure measurement / modelling
● Compartmentalization*
● Fluid analysis

Resource Initial Reservoir Modelling:

● Petrophysical analysis
Management
● Well correlation
● Facies identification
● Facies mapping
● Property mapping
● Resource / reserve estimation

Reservoir model Update

Reservoir property Estimation:
3D Facies Mapping:
● Rock typing
● J-function / Saturation Height Model

Portfolio ● Reservoir Mapping

● Upside Potential Exploration Workflow
Management
● LRLC
● Bypassed Zone

Automation Candidate, tech acquired

Automation Candidate, tech feasible

epo@[Link]
 USE WITH CAUTION!!
THIS IS NOT WORK OF MAGIC:

Always have Geologist check the Results..!

Goal Description
1. To automate routine geological
Geologist
work that are repetitive, very prone
to subjectivity, and require a lot of
time to work

2. To provide ‘pre-interpreted maps’

for geologist to work on
interpretation
Discrete
Redraw the CRS
Raw Data QC Processed QC CRS MAPS Overlay
3. To design a workflow that can be MAP
Information
replicated in other field/basin

4. To train Geologist from SKK migas

to understand how to do EDA, and
to do simple Automation Task

epo@[Link]
 Algorithm: Automated Well Correlation
Target Well
Automatic Correlation:
Source Well

[Link]
epo@[Link]
 Algorithm: Facies Mapping

ore

ore
Automatic 2D Facies Mapping:

Sh

Sh
e

e
fac

fac
ore

ore
Sh

Sh
ore

ore
fsh

fsh
Of

Of
[Link]
mapping-depositional-facies-belt-wit
h-svm-in-python-some-improvization
-4a92d91ba2ec

epo@[Link]
 Algorithm: Facies Mapping
Automatic 3D Facies Mapping:

[Link]
epo@[Link]
 Non
Hydrocarbon
Exploration and
Development
Use Cases

epo@[Link]
 Gawler Project: Mineral Deposit
Prediction
Goal statement: This is what Noisy Data looks like..!!
- Make estimation of selected mineral
deposit using Data Science Delustering? Reclustering? Normalization?

Data:
- Remote sensing (Sentinel)
- Regional Gravity
- Regional Magentic +50gb of Raster data
- Regional Radiometric
- Mineral deposit location
- Geochemistry data 10gb of geochem data

Problem we dealt with:

- BIG DATA
- Multi format (raster vs vector)
- Data noise

epo@[Link]
 Raw Data: Vector Cu Deposit Distribution

Gawler Project: Mineral Deposit

Prediction
Raw Data: Vector Cu Deposit Distribution Prediction

Raw Data: Raster

Raw Data: Radiometric data

Results:
- Prediction Result: Fairly good Prediction

Raw Data: Gravity Data

epo@[Link]
 Lineament Detection using DEM
Raw Data: SRTM Lineament Mapping KDE Map of Lineament Density

epo@[Link]
 Lithology Prediction using
Hyperspectral Image
GRDC mAp
KNN SVM SVM + sPATIAL

epo@[Link]
 Styrene biomarker data analysis

[Link to Post]

epo@[Link]
 Kriging using open-source tools

[Link to Post]

epo@[Link]
 SedStrat off Office Hour
Informal Random talk about Data
science/machine learning / Deep
learning and its application in
Geoscience

Email us at:
epo@[Link]
[Link]@[Link]
aviandito@[Link]
yogipamadya@[Link]

[Link]

epo@[Link]
 Thank
you

epo@[Link]